fix: implement product line seasonality for forecast fallback

services/dataProcessor.ts

@@ -1768,14 +1768,49 @@ export const calculateForecastViewData = (
         monthMap.set(m, (monthMap.get(m) || 0) + r.units);
     });
 
+    // 1b. Determine Line-Level Weights (NEW STRATEGY)
+    const lineWeightsMap = new Map<string, number[]>();
+    const linesMap = new Map<string, SalesRecord[]>();
+
+    historicalData.forEach(r => {
+        if (!r.line) return;
+        if (!linesMap.has(r.line)) linesMap.set(r.line, []);
+        linesMap.get(r.line)!.push(r);
+    });
+
+    linesMap.forEach((records, line) => {
+        // We use the same getWeightsInfo logic but for the whole line
+        const info = getWeightsInfo(records);
+        if (info) {
+            lineWeightsMap.set(line, info.weights);
+        } else {
+            // Fallback for line if it has data but odd distribution? 
+            // Actually getWeightsInfo returns null only if total=0.
+            // If we have records but 0 units total, we skip map set, so it will fall to global.
+        }
+    });
+
     return forecastData.map(fc => {
         const identifier = fc.asin.toUpperCase();
         const meta = asinMetadata.get(identifier);
+
+        // Resolve Line: Try meta first, then forecast file
+        const resolvedLine = meta?.line || fc.line || "Unassigned";
+
         const avgWeeklySales = velocityMap?.get(identifier) || 0;
 
         // 2. Determine weights for this ASIN
         const productHistoricalRecords = dataByAsinHistorical.get(identifier) || [];
-        let productWeights = globalWeights;
+
+        // LAYERED FALLBACK STRATEGY:
+        // Level 1: Product's own history (Most accurate)
+        // Level 2: Product Line's history (Good for new items in known category e.g. Advent Calendars)
+        // Level 3: Global/Pan-EU history (Generic fallback)
+
+        const lineWeights = lineWeightsMap.get(resolvedLine);
+        const baselineWeights = lineWeights || globalWeights;
+
+        let finalWeights = baselineWeights;
 
         if (productHistoricalRecords.length > 0) {
             const historyToUse = isUkOnly
@@ -1785,12 +1820,9 @@ export const calculateForecastViewData = (
             const info = getWeightsInfo(historyToUse);
             if (info) {
                 // Adaptive Blending (Bayesian Shrinkage):
-                // We blend local seasonality with global seasonality based on how many months of data we have.
-                // 12 months = 85% local, 15% global (safety net)
-                // 6 months = 42.5% local, 57.5% global
-                // 0 months = 0% local, 100% global
+                // We blend local seasonality with baseline (Line or Global) based on data density.
                 const trustFactor = (info.monthsCount / 12) * 0.85;
-                productWeights = info.weights.map((w, i) => (w * trustFactor) + (globalWeights[i] * (1 - trustFactor)));
+                finalWeights = info.weights.map((w, i) => (w * trustFactor) + (baselineWeights[i] * (1 - trustFactor)));
             }
         }
 
@@ -1800,7 +1832,7 @@ export const calculateForecastViewData = (
         let totalForecastUnits = 0;
 
         MONTH_ORDER.forEach((m, idx) => {
-            const forecastUnits = Math.round(fc.annualForecast * productWeights[idx]);
+            const forecastUnits = Math.round(fc.annualForecast * finalWeights[idx]);
             const actualUnits = actuals2026.get(identifier)?.get(m) || 0;
 
             monthlyData[m] = {